Gameweek 10 Verdict: Our AI Scored 5/10 and Two Results Fooled All Four Forecasters

Published 10 November 2026 · FootyPulse AI Analysis

5 min read

Gameweek 10 Verdict: Our AI Scored 5/10 and Two Results Fooled All Four Forecasters

FootyPulse's AI called 5 of 10 Premier League results correctly in Gameweek 10: a point behind BBC's AI and the BBC guest (both 6/10), and a point ahead of Chris Sutton (4/10). It was a mid-table finish in a round where home sides won six times.

Every pick here is on the record. BBC's predictions are published ahead of the gameweek, and FootyPulse's AI predictions are locked 12 hours before kick-off and do not change afterwards. All results are scored on 90-minute outcomes (home win, draw or away win).

Correction: our Friday preview carried earlier probabilities for Ipswich v Bournemouth and Crystal Palace v Liverpool. The model updated before the 12-hour lock. This review uses the locked figures, which are the ones scored.

The Gameweek 10 scoreboard

ForecasterCorrectAccuracy
BBC's AI6/1060%
BBC guest6/1060%
FootyPulse AI5/1050%
Chris Sutton4/1040%

One gameweek is ten matches. A single result moves any forecaster by ten percentage points, so this table shows who had a good weekend, not who is the better forecaster. For the season-long record, see the FootyPulse accuracy page and the live AI vs Pundits tracker. The match-by-match summary is on the Gameweek 10 verdict page, and our pre-round thinking is in the Gameweek 10 predictions preview.

Bar chart comparing Gameweek 10 accuracy: BBC's AI 6/10, BBC guest 6/10, FootyPulse AI 5/10, Chris Sutton 4/10.

How close was it, really?

Closer than the table suggests:

  • FootyPulse's AI and BBC's AI made the same pick in eight of ten matches. The only differences were Ipswich v Bournemouth (we picked the home win, BBC's AI the draw; both were wrong) and Manchester United v Aston Villa. BBC's AI called the draw at Old Trafford, and that single match is the whole gap between us.
  • The BBC guest's 6/10 came partly from two calls nobody else made: a Leeds win over Tottenham and a Bournemouth win at Ipswich. The guest also missed four.
  • Chris Sutton's 4/10 included a miss at Fulham, where the other three forecasters all called the draw.

Those are the differences between the picks, not evidence that anyone has a better method. See the BBC's published Gameweek 10 predictions.

Two results beat all four forecasters

Three matches were called correctly by all four: Everton 3-1 Coventry City, Arsenal 1-0 Hull City and Crystal Palace 1-3 Liverpool. At the other extreme, two matches beat everybody.

Both were home wins by the side the forecasters rated as the underdog. They are also our two lowest-probability results of the weekend.

What we got wrong

Grid of all ten Gameweek 10 Premier League fixtures showing results, FootyPulse's pick, correct or wrong verdict, and how many of four forecasters were right.

FootyPulse's AI missed five of ten. Here is every miss, with the probability the model gave to the result that actually happened.

MatchOur pick (probability)ResultProbability we gave the actual outcome
Nottingham Forest 3-1 Manchester CityManchester City (58%)Forest win18%
Sunderland 1-0 ChelseaChelsea (55%)Sunderland win18%
Leeds United 1-0 Tottenham HotspurTottenham (42%)Leeds win28%
Ipswich Town 1-2 BournemouthIpswich (42%)Bournemouth win30%
Manchester United 4-4 Aston VillaAston Villa (41%)Draw31%

Bar chart of the probability FootyPulse's AI gave to the actual result in its five Gameweek 10 misses; all ten matches averaged 42%.

Three patterns stand out:

  • The away leans struggled. We backed an away side in five matches and only one landed (Liverpool). Tottenham, Manchester City, Chelsea and Aston Villa did not win.
  • Home sides won six of ten. Our probabilities implied about 3.8 home wins across the ten matches. That is a gap, but ten matches is far too few to call it a bias.
  • Only two misses were high-conviction. Forest v City (58%) and Sunderland v Chelsea (55%) were our two most confident misses. The other three came with a top pick of 42% or lower, and at Old Trafford the draw we missed was given 31%.

Across all ten matches, the model gave the eventual result an average probability of 42%. For a three-way outcome with no information, you would expect 33%. That is useful context, not proof of skill from one round.

Did the half-time call help?

FootyPulse's half-time re-predictions are shown alongside the pre-match call on the accuracy page. For Gameweek 10:

CallCorrectExact scores
Pre-match call5/101
Half-time call6/103

Bar chart: pre-match call 5 of 10 correct, half-time call 6 of 10 correct.

The difference is one match: Ipswich v Bournemouth, where the half-time call switched to Bournemouth and was right. The same four other matches were wrong at both stages. One match is not a trend, so this is too early to tell. We will report it as the sample grows. You can read how the half-time update works on the methodology page.

What it means

  • No clear winner among the forecasters. All four finished within two matches of each other, and the table could have reordered on one result.
  • The model's call was reasonable but not right. Five hits against about 5.4 expected from its own top-pick probabilities (the sum of the ten top picks is 5.36). In other words, the model did roughly what it said it would do.
  • The thing to watch is whether the away-lean pattern repeats. If it does across a larger sample, that is a calibration issue worth publishing.

Gameweek 10 was the last round before the international break. We will track every pick, including the misses, on the AI vs Pundits page. Which result surprised you most: Forest over City or Sunderland over Chelsea?

FAQ

How accurate was FootyPulse's AI in Premier League Gameweek 10? FootyPulse's AI called 5 of 10 results correctly (50%). BBC's AI and the BBC guest each got 6 of 10, and Chris Sutton got 4 of 10.

Which Gameweek 10 results did every forecaster get wrong? Nottingham Forest 3-1 Manchester City and Sunderland 1-0 Chelsea. FootyPulse's AI, BBC's AI, Chris Sutton and the BBC guest all missed both.

When are FootyPulse's AI predictions locked? FootyPulse's AI predictions are locked 12 hours before kick-off and do not change afterwards. BBC picks are published ahead of the gameweek.

Do half-time re-predictions improve accuracy? In Gameweek 10 the half-time call was right in 6 of 10 matches against 5 of 10 for the pre-match call, a difference of one match. That is too small a sample to draw a conclusion.

Share this article: